July 22, 2026, (Inside AI) — British publisher Bloomsbury Publishing confirmed on Tuesday it will receive compensation from a landmark $1.5 billion settlement with AI startup Anthropic. The deal resolves claims the company used copyrighted books to train its Claude language models without permission.
The settlement, finalized in a U.S. federal court, marks one of the largest copyright resolutions in the generative AI era. It stems from a class-action lawsuit filed by authors and publishers who alleged Anthropic scraped thousands of books from pirate sites. The sum will be distributed among rights holders, with Bloomsbury among the named beneficiaries.
Anthropic did not admit liability but agreed to the payout to avoid prolonged litigation. A company spokesperson stated: "We respect intellectual property and are committed to working with creative industries to build a sustainable AI ecosystem."
The Battle Over Training Data Heats Up #
This case is part of a broader reckoning over how AI firms source training data. Unlike earlier lawsuits against OpenAI and Meta, the Anthropic suit focused on the use of Books3, a dataset containing 196,640 pirated books. Researchers have shown that such datasets underpin many large language models.
The settlement sets a financial precedent. While Getty Images settled with Stability AI for an undisclosed sum, the Anthropic deal dwarfs previous payouts. It may pressure other AI developers to license content proactively or face similar legal exposure.
Legal experts note the outcome avoids a court ruling on fair use. The U.S. Copyright Office has yet to issue definitive guidance on AI training, leaving the doctrine in limbo. This settlement sidesteps that uncertainty, offering cash instead of clarity.
Publishers Secure a Win, but Questions Linger #
For Bloomsbury, the windfall is a timely boost. The publisher of Harry Potter and academic works has invested in digital infrastructure, but physical sales remain dominant. The settlement could fund further digitization or author advances. However, the deal does not cover future use. Anthropic can still train on copyrighted works if it licenses them. The settlement includes a mechanism for ongoing royalty payments if the company continues using the disputed data.
Author groups remain skeptical. Mary Rasenberger, CEO of the Authors Guild, said: "A one-time payout doesn’t fix the systemic problem. We need transparency and opt-out mechanisms for all AI training."
Meanwhile, regulators are watching. The U.K. Intellectual Property Office is drafting a code of practice on AI and copyright, while the EU AI Act mandates disclosure of training data sources. These moves could render such settlements less common in the future.
Anthropic, valued at $60 billion, can absorb the cost. The company recently raised $3.5 billion in funding and competes fiercely with OpenAI and Google DeepMind. Settling may be a strategic move to clear legal hurdles ahead of a potential IPO.
The settlement also highlights the growing power of collective action. The class-action model allowed individual authors to band together, overcoming the resource gap with tech giants. Similar suits against Nvidia and Microsoft are pending.
As AI models grow more voracious for data, the tension between innovation and creators' rights will only intensify. This settlement offers a temporary truce, but the war over data is far from over.